Claude's 5-hour session limit, in plain English

Claude's 5-hour session limit is a rolling window, not a fixed daily slot. It starts with your first message of each cycle, and the budget refills exactly 5 hours later - even if you used it all up in the first hour. Once you understand the rolling math, both the wait and the workaround become predictable.

Update, 2026-09-19: Max plans no longer run a separate weekly limit scoped to Opus. Opus and Sonnet are both current, fully available models, and both draw from the single weekly limit that covers every model. The second weekly meter on Max is scoped to Fable instead: Max includes Fable for up to 50% of the weekly limit, after which Fable keeps running on pay-as-you-go usage credits. Pro has one weekly limit across all models and bills Fable from usage credits from the first message. Checked against a live Claude Max 5x account that used Fable, Opus and Sonnet in the same week: Settings > Usage shows current session, this week, and Fable this week, with no Opus meter.

The short answer

What the 5-hour window actually is

Per Anthropic's help-center article on usage limits, the 5-hour session cap is one of three rolling limits on paid Claude plans. It is the shortest - the others are a weekly cap covering every model, and on Max a second weekly meter scoped to Fable. All three use the same rolling-window mechanic: the clock starts at your first message of the cycle, not at midnight or on any fixed schedule.

The session cap is a budget inside the window, not simply a time limit. Think of it like a data plan that renews every 5 hours. You can use it all at once, or stretch it over the full 5 hours - either way the window expires 5 hours after it opened, and the next message after expiry opens a fresh cycle.

The rolling-window math

This is the part that trips most people. Because the window is rolling and anchored on your first message, two users on the same plan can have completely different reset times.

The key insight: you never wait the full 5 hours unless you exhausted the budget in the very first minute. If you sent your first message at 10:00 and ran out of budget at 13:30, the window expires at 15:00 - a 1.5-hour wait, not 5. The earlier you deplete the budget, the longer you wait; the later you deplete it, the shorter the remaining window.

The banner Claude shows when you hit the cap gives you the exact expiry time in your local timezone. You do not have to do the arithmetic yourself - but understanding the rolling model helps you predict it before you hit it, which is where tools like tokn.watch come in.

What counts against the 5-hour session cap

Every token your account consumes counts, regardless of which surface generated it:

• Chat messages on claude.ai - both the text you send and the text Claude returns.
• Claude Code tool calls - every file read, glob, bash command, and edit Claude executes burns from the same pool.
• Claude desktop and mobile app messages.
• Project context loaded automatically at the start of each session - large project instructions are a real budget item.

This shared-pool design is documented in Anthropic's usage limits article and confirmed by the pattern most Claude Code users discover the hard way: a long agentic run on a large codebase can exhaust the 5-hour budget in under an hour because each tool call adds up. The full breakdown of how all three limits interact is in how Claude's usage limits really work in 2026.

What does not reset the 5-hour cap

If you have hit the cap, none of these unlock it:

• Logging out and back in. The cap is account-level and server-side.
• Starting a new conversation. Caps are per account, not per chat.
• Switching from claude.ai to Claude Code or vice versa. Same pool.
• Switching models. The 5-hour session cap is shared across all models - Sonnet, Haiku, and Opus all draw from the same window. (Switching models helps only when you have reached Fable's 50% ceiling on Max, which is a different limit.)
• Refreshing the page or restarting the app.
• Waiting until midnight. The window is anchored on your first message, not on clock time. Midnight is irrelevant.

The only mechanism that resets the 5-hour session cap is time: the 5-hour window completing from the point of your first message.

How to know when your session resets

Three ways, in order of friction:

1. The banner. When Claude blocks your message, the banner shows the reset time in your local timezone. This is the most precise signal but arrives after you are already locked out.
2. Settings > Usage on claude.ai. Shows each cap's percentage consumed and the reset timestamp. Useful for checking mid-session before you hit the wall.
3. /usage in Claude Code. Same numbers as the settings page, in your terminal. Per Anthropic's Claude Code monitoring docs, this is the canonical in-CLI view of your current budget.

How the 5-hour cap behaves in Claude Code specifically

Claude Code users hit the 5-hour cap proportionally more often than chat users, because agentic tasks burn tokens at a much higher rate. A single Claude Code session doing a large refactor - reading dozens of files, running tests, writing diffs - can consume the full session budget in 45 minutes, leaving you with 4 hours and 15 minutes of lockout.

The Register reported in March 2026 that Anthropic tightened Claude Code limits, and the community discussion that followed (including a 30-plus issue thread on the anthropics/claude-code GitHub) confirmed that tool-heavy agentic sessions are by far the fastest way to exhaust the 5-hour window.

Practical Claude Code tactics when the 5-hour budget is low:

• Run /compact before starting a big task. Compressing the conversation context reduces what Claude has to re-read on each turn, and lower re-read cost means more tasks per window. The full list of tactics is in 13 ways to save Claude tokens.
• Commit early. The lockout is a natural commit boundary. Uncommitted work is safe on disk but you want the branch state recoverable.
• Check /usage before starting a big refactor. If you are at 70% of the 5-hour budget, start the task after the reset rather than mid-window.
• Use worktrees for parallel tasks only when you have a fresh window - a 25% window shared across two worktrees depletes twice as fast.

Session cap vs. weekly cap: they are separate

A common point of confusion: hitting the 5-hour session cap does not mean you have hit your weekly limit. They are two independent budgets on different timers.

You can hit the 5-hour session cap several times in a single week without touching the weekly all-models cap. Conversely, hitting the weekly cap does not affect the session timer - it simply means there is no budget in either window until the weekly one completes. The interaction between all three limits is covered in full in when does Claude reset?, and Fable models sit on top of the same weekly pool with their own ceiling - see does Claude Pro include Fable 5.1? for how that works per plan.

Does your plan change the 5-hour window?

The window length (5 hours) is the same across Pro, Max 5x, and Max 20x. What differs is the size of the budget inside each window - Max plans get proportionally larger session budgets, which is why they take longer to deplete before hitting the cap.

Anthropic does not publish the exact budget sizes in usage units; the only visible metric is the percentage in Settings > Usage. But the practical effect is observable: a Max 20x user doing the same set of tasks will hit the 5-hour cap less often than a Pro user, because the window is wider in absolute capacity even though it is the same 5 hours long.

"Usage limits are designed to ensure fair access for all Claude users. We do not publish the absolute token counts for each window because they vary by model and can change as we improve the system." Anthropic, usage limits help center (paraphrased)

Frequently asked questions